Smart Pricing and Replenishment Decision Making Based on Machine Learning Algorithms

In the management of fresh food supermarkets, making optimal replenishment decisions for various vegetable categories daily is crucial when the specific items and purchase prices are not accurately known. To assist supermarkets in selecting reasonable sales combinations, this article formulates intelligent pricing and replenishment strategies for vegetable products using clustering, random forest models, and machine learning algorithms for modeling and problem solving. The main research contents are as follows: For problem one, data preprocessing is conducted initially. The distribution patterns of various categories are explored using wavelet analysis, leading to the conclusion that most data exhibit strong periodicity. The interrelationships among various categories are determined using the elbow method to identify the optimal parameters, followed using K-means clustering and hierarchical clustering to roughly divide the sales volumes of various categories into two classes. Upon finding that the data does not conform to a normal distribution, the Spearman correlation coefficient analysis method is applied for data processing. To analyze the distribution patterns of individual vegetable products, seasonal decomposition is used to analyze their periodic characteristics and certain anomalies. The interrelationships among individual vegetable products are evaluated using the Granger causality test, which shows that there is a significant impact on the sales volumes of most vegetable products, with only a small portion of the data showing weaker sales volume impacts between products. For problem two, a weighted sales unit price model is established based on the differences in sales unit prices of individual vegetable products. This is followed by constructing an XGBoost machine learning algorithm embedded with grid search to predict the sales volumes of various vegetable categories in the coming week, and the model is evaluated for six categories, showing excellent fitting effects. Subsequently, an LSTM-RF time series forecasting model is built to predict the cost of various vegetable categories in the coming week. An objective function is established with the aim of maximizing supermarket profits, and the optimal daily replenishment total and pricing strategy are obtained using a genetic algorithm. For problem three, the category problem is further subdivided into the problem of daily replenishment quantity and pricing strategy for individual products based on the work conducted for problem two. Some modifications are made to the weighted sales unit price model, and a grid search-based XGBoost random forest prediction model and LSTM time series prediction model are continued to be established to predict the daily sales volume and cost price of vegetable products. The optimal daily replenishment quantity and pricing strategy for individual products are then determined using the particle swarm optimization algorithm. For problem four, the multiple factors considered in the models established for problems two and three are addressed. Relevant data collection is proposed to address the shortcomings of the models, such as seasonal factors, promotional factors, and market factors. Furthermore, model optimization and parameter adjustments are made to enhance the flexibility of the model.